This PDF is a selection from a published volume from the National Bureau of Economic Research
Volume Title: International Dimensions of Monetary Policy Volume Author/Editor: Jordi Gali and Mark J. Gertler, editors Volume Publisher: University of Chicago Press
Volume ISBN: 0-226-27886-7
Volume URL: http://www.nber.org/books/gert07-1 Conference Date: June 11-13, 2007
Publication Date: February 2010
Chapter Title: The Macroeconomic Effects of Oil Price Shocks: Why are the 2000s so different from the 1970s?
Chapter Author: Olivier J. Blanchard, Jordi Galí
Chapter URL: http://www.nber.org/chapters/c0517
Chapter pages in book: (373 - 421)
373
7.1 Introduction
Since the 1970s, and at least until recently, macroeconomists have viewed changes in the price of oil as an important source of economic fl uctuations, as well as a paradigm of a global shock, likely to a ffect many economies simultaneously. Such a perception is largely due to the two episodes of low growth, high unemployment, and high infl ation that characterized most industrialized economies in the mid and late 1970s. Conventional accounts of those episodes of stagfl ation blame them on the large increases in the price of oil triggered by the Yom Kippur war in 1973, and the Iranian revolution of 1979, respectively.
1The events of the past decade, however, seem to call into question the rele- vance of oil price changes as a signifi cant source of economic fl uctuations.
The reason: since the late 1990s, the global economy has experienced two oil shocks of sign and magnitude comparable to those of the 1970s but, in contrast with the latter episodes, both gross domestic product (GDP) growth
7
The Macroeconomic E ffects of Oil Price Shocks
Why Are the 2000s so Di fferent from the 1970s?
Olivier J. Blanchard and Jordi Galí
Olivier J. Blanchard is the Class of 1941 Professor at the Massachusetts Institute of Tech- nology and a research associate of the National Bureau of Economic Research. Jordi Galí is the Director of the Center for Research in International Economics (CREI), a professor of economics at the Universitat Pompeu Fabra, and a research associate of the National Bureau of Economic Research. We are grateful for helpful comments and suggestions to Julio Rotemberg, John Parsons, Lutz Kilian, José de Gregorio, Gauti Eggertsson, Carlos Montoro, Tomaz Cajner, and participants at NBER ME Meeting, the NBER conference on International Dimensions of Monetary Policy, and seminars at CREI- UPF, LSE, GIIS (Geneva), Fundación Rafael del Pino, and Bank of Portugal. We thank Davide Debortoli for excellent research assistance, and the NSF and the Banque de France Foundation for fi nancial assistance. This version updates an earlier version, using data up to 2007:3 and corrected oil shares.
1. Most undergraduate textbooks make an unambiguous connection between the two oil price hikes of 1973 and 1974 and 1979 and 1980 and the period of stagfl ation that ensued. (See, e.g., Mankiw [2007, 274].)
374 Olivier J. Blanchard and Jordi Galí
and infl ation have remained relatively stable in much of the industrialized world.
Our goal in this chapter is to shed light on the nature of the apparent changes in the macroeconomic e ffects of oil shocks, as well as on some of their possible causes. Disentangling the factors behind those changes is obviously key to assessing the extent to which the episodes of stagfl ation of the 1970s can reoccur in response to future oil shocks and, if so, to under- standing the role that monetary policy can play in order to mitigate their adverse e ffects.
One plausible hypothesis is that the e ffects of the increase in the price of oil proper have been similar across episodes, but have coincided in time with large shocks of a very di fferent nature (e.g., large rises in other commod- ity prices in the 1970s, high productivity growth, and world demand in the 2000s). That coincidence could signifi cantly distort any assessment of the impact of oil shocks based on a simple observation of the movements in aggregate variables around each episode.
In order to evaluate this hypothesis one must isolate the component of macroeconomic fl uctuations associated with exogenous changes in the price of oil. To do so, we identify and estimate the e ffects of an oil price shock using structural Vector Autoregression (VAR) techniques. We report and compare estimates for di fferent sample periods and discuss how they have changed over time. We follow two alternative approaches. The fi rst one is based on a large VAR, and allows for a break in the sample in the mid- 1980s.
The second approach is based on rolling bivariate VARs, including the price of oil and one other variable at a time. The latter approach allows for a gradual change in the estimated e ffects of oil price shocks, without imposing a discrete break in a single period.
Two conclusions clearly emerge from this analysis: fi rst, there were indeed other adverse shocks at work in the 1970s; the price of oil explains only part of the stagfl ation episodes of the 1970s. Second, and importantly, the e ffects of a given change in the price of oil have changed substantially over time.
Our estimates point to much larger e ffects of oil price shocks on infl ation and activity in the early part of the sample; that is, the one that includes the two oil shock episodes of the 1970s.
Our basic empirical fi ndings are summarized graphically in fi gure 7.1 (we postpone a description of the underlying assumptions to section 7.3).
The left- hand graph shows the responses of U.S. (log) GDP and the (log)
consumer price index (CPI) to a 10 percent increase in the price of oil,
estimated using pre- 1984 data. The right- hand graph displays the corre-
sponding responses, based on post- 1984 data. As the fi gure makes clear,
the response of both variables has become more muted in the more recent
period. As we show following, that pattern can also be observed for other
variables (prices and quantities) and many (though not all) other countries
considered. In sum, the evidence suggests that economies face an improved
The Macroeconomic Effects of Oil Price Shocks 375
trade- o ff in the more recent period, in the face of oil price shocks of a similar magnitude.
We then focus on the potential explanations for these changes over time.
We consider three hypotheses, not mutually exclusive. First, real wage rigidi- ties may have decreased over time. The presence of real wage rigidities gen- erates a trade- o ff between stabilization of infl ation and stabilization of the output gap. As a result, and in response to an adverse supply shock and for a given money rule, infl ation will generally rise more and output will decline more, the slower real wages adjust. A trend toward more fl exible labor mar- kets, including more fl exible wages, could thus explain the smaller impact of the more recent oil shocks.
Second, changes in the way monetary policy is conducted may be respon- sible for the di fferential response of the economy to the oil shocks. In par- ticular, the stronger commitment by central banks to maintaining a low and stable rate of infl ation, refl ected in the widespread adoption of more or less explicit infl ation targeting strategies, may have led to an improvement in the policy trade- o ff that would make it possible to have a smaller impact of a given oil price increase on both infl ation and output simultaneously.
Third, the share of oil in the economy may have declined su fficiently since the 1970s to account for the decrease in the e ffects of its price changes. Under that hypothesis, changes in the price of oil have increasingly turned into a sideshow, with no signifi cant macroeconomic e ffects (not unlike fl uctuations in the price of caviar).
To assess the merits of the di fferent hypotheses we proceed in two steps.
First, we develop a simple version of the new- Keynesian model where
(imported) oil is both consumed by households and used as a production
input by fi rms. The model allows us to examine how the economy’s response
to an exogenous change in the price of oil is a ffected by the degree of real
Fig. 7.1 U.S.—Impulse response to an oil price shock376 Olivier J. Blanchard and Jordi Galí
wage rigidities, the nature and credibility of monetary policy, and the share of oil in production and consumption. We then look for more direct evi- dence pointing to the relevance and quantitative importance of each of those hypotheses. We conclude that all three are likely to have played an important role in explaining the di fferent effects of oil prices during the 1970s and during the last decade.
The chapter is organized as follows. Section 7.1 gives a short summary of how our chapter fi ts in the literature. Section 7.2 presents basic facts. Sec- tion 7.3 presents results from multivariate VARs. Section 7.4 presents results from rolling bivariate VARs. Section 7.5 presents the model. Section 7.6 uses the model to analyze the role of real rigidities, credibility in monetary policy, and the oil share. Section 7.7 concludes.
7.1 Relation to the Literature
Our chapter is related to many strands of research. The fi rst strand is concerned with the e ffects of oil price shocks on the economy. The seminal work in that literature is Bruno and Sachs (1985), who were the fi rst to ana- lyze in depth the e ffects of oil prices of the 1970s on output and infl ation in the major industrialized countries. They explored many of the themes of our chapter, the role of other shocks, the role of monetary policy, and the role of wage setting.
On the empirical side, Hamilton showed in a series of contributions (see, in particular, Hamilton [1983, 1996]) that most of U.S. recessions were pre- ceded by increases in the price of oil, suggesting an essential role for oil price increases as one of the main causes of recessions. The stability of this relation has been challenged by a number of authors, in particular Hooker (1996). Our fi nding that the e ffects of the price of oil have changed over time is consistent with the mixed fi ndings of this line of research.
On the theoretical side, a number of papers have assessed the ability of
standard models to account for the size and nature of the observed e ffects
of oil price shocks. Thus, Rotemberg and Woodford (1996) argued that it
was di fficult to explain the sheer size of these effects in the 1970s. They
argued that something else was going on; namely, an endogenous increase
in the markup of fi rms, leading to a larger decrease in output. Finn (2000)
showed that e ffects of the relevant size could be generated in a perfectly
competitive real business cycle (RBC) model, by allowing for variable capi-
tal utilization. Neither mechanism would seem to account for the depth of
the e ffects of the 1970s and not in the 2000s. The latter observation moti-
vates our focus on the role of real wage rigidities, and the decline in these
rigidities over time, an explanation we fi nd more convincing than changes in
either the behavior of markups or capacity utilization over time. In follow-
ing this line, we build on our earlier work on the implications of real wage
The Macroeconomic Effects of Oil Price Shocks 377
rigidities and their interaction with nominal price stickiness (Blanchard and Galí 2007).
A second strand of research related to the present chapter deals with the possible changes over time in the e ffects of oil shocks. Of course, that strand is in turn related to the literature on the “Great Moderation,” a term used to refer to the decrease in output fl uctuations over the last thirty years (e.g., Blanchard and Simon 2001; Stock and Watson 2003). The latter literature has tried to assess to what extent the declines in volatility have been due to
“good luck” (i.e., smaller shocks) or changes in the economy’s structure (including policy changes). In that context, some authors have argued that the stagfl ations of the 1970s were largely due to factors other than oil. Most prominently, Barsky and Kilian (2002) argue that they may have been partly caused by exogenous changes in monetary policy, which coincided in time with the rise in oil prices. Bernanke, Gertler, and Watson (1997) argue that much of the decline in output and employment was due to the rise in interest rates, resulting from the Fed’s endogenous response to the higher infl ation induced by the oil shocks.
While our evidence suggests that oil price shocks can only account for a fraction of the fl uctuations of the 1970s, our fi ndings that the dynamic e ffects of oil shocks have decreased considerably over time, combined with the observation that the oil shocks themselves have been no smaller, is con- sistent with the hypothesis of structural change.
We know of four papers that specifi cally focus, as we do, on the chang- ing impact of oil shocks. Hooker (2002) analyzes empirically the changing weight of oil prices as an explanatory variable in a traditional Phillips curve specifi cation for the U.S. economy. He fi nds that pass- through from oil to prices has become negligible since the early 1980s, but cannot fi nd evidence for a signifi cant role of the decline in energy intensity, the deregulation of energy industries, or changes in monetary policy as a factor behind that lower pass- through. De Gregorio, Landerretche, and Neilson (2007) provide a variety of estimates of the degree of pass- through from oil prices to infl a- tion, and its changes over time, for a large set of countries. In addition to esti- mates of Phillips curves along the lines of Hooker (2002), they also provide evidence based on rolling VARs, as we do in the present chapter, though they use a di fferent specifi cation and focus exclusively on the effects on infl ation.
Their paper also examines a number of potential explanations, including a
change in the response of the exchange rate (in the case of non- U.S. coun-
tries), and the virtuous e ffects of being in a low infl ation environment. In
two recent papers, developed independently, Herrera and Pesavento (2007)
and Edelstein and Kilian (2007) also document the decrease in the e ffects of
oil shocks on a number of aggregate variables using a VAR approach. Her-
rera and Pesavento, following the approach of Bernanke, Gertler, and Wat-
son (1997), explore the role of changes in the response of monetary policy
378 Olivier J. Blanchard and Jordi Galí
to oil shocks in accounting for the more muted e ffects of those shocks in the recent period. Their answer is largely negative: their fi ndings point to a more stabilizing role of monetary policy in the 1970s relative to the recent period. Edelstein and Kilian focus on changes in the composition of U.S.
automobile production, and the declining importance of the U.S. automo- bile sector. Given that the decline in the e ffects of the price of oil appears to be present in a large number of Organization for Economic Cooperation and Development (OECD) countries, this explanation appears perhaps too U.S.- specifi c.
7.2 Basic Facts
Figure 7.2 displays the evolution of the price of oil since 1970. More specifi cally, it shows the quarterly average price of a barrel of West Texas Intermediate, measured in U.S. dollars.
2The fi gure shows how a long spell of stability came to an end in 1973, triggering a new era characterized by large and persistent fl uctuations in the price of oil, punctuated with occasional sharp run- ups and spikes, and ending with the prolonged rise of the past few years. The shaded areas in the fi gure correspond to the four large oil shock episodes discussed following.
Figure 7.3 displays the same variable, now normalized by the U.S. GDP defl ator, and measured in natural logarithms. This transformation gives us a better sense of the magnitude of the changes in the real price of oil. As the fi gure makes clear, such changes have often been very large, and concen- trated over relatively short periods of time.
It is useful to start with descriptive statistics associated with the large oil shocks visible in the previous fi gures. We defi ne a large oil shock as an epi- sode involving a cumulative change in the (log) price of oil above 50 percent, sustained for more than four quarters. This gives us four episodes, starting in 1973, 1979, 1999, and 2002, respectively. Exact dates for each run- up are given in table 7.1 (given our defi nition, the largest price changes need not coincide with the starting date, and, indeed, they do not). For convenience we refer to those episodes as O1, O2, O3, and O4, respectively. Note that this criterion leaves out the price rise of 1990 (triggered by the Gulf War), due to its quick reversal. We also note that O3 is somewhat di fferent, since it is preceded by a signifi cant price decline.
Table 7.1 lists, for each episode: (a) the run- up period; (b) the date at which the cumulative log change attained the 50 percent threshold (which we use as a benchmark date in the following); and (c) the percent change from trough to peak (measured by the cumulative log change), both in nominal
2. The description of the stylized facts discussed following is not altered signifi cantly if one uses alternative oil price measures, such as the PPI index for crude oil (used, e.g., by Hamilton [1983] and Rotemberg and Woodford [1996]) or the price of imported crude oil (e.g., Kilian 2006).
Fig. 7.2 Oil price ($ per barrel)
Fig. 7.3 Log real oil price
380 Olivier J. Blanchard and Jordi Galí
and real terms. The duration of the episodes ranges from three quarters (O1) to twenty quarters (O4).
3Interestingly, the size of the associated nominal price rise is roughly similar across episodes—around 100 percent. A similar characteriza- tion emerges when we use the cumulative change in the real price of oil (with the price normalized by the GDP defl ator), except for O2, where the rise is somewhat smaller because of the high rate of infl ation during that episode. In short, the four episodes involve oil shocks of a similar mag- nitude. In particular, the numbers do not seem to justify a characteriza- tion of the two recent shocks as being milder in size than the shocks of the 1970s.
In spite of their relatively similar magnitude, these four oil shock episodes have been associated with very di fferent macroeconomic performances. Fig- ures 7.4 and 7.5, which show, respectively, the evolution of (annual) CPI infl ation and the unemployment rate in the United States over the period 1970:1 through 2007:3, provide a visual illustration.
Each fi gure shows, in addition to the variable displayed, the (log) real price of oil and the four shaded areas representing our four oil shock epi- sodes. Note that the timing of O1 and O2 coincide with a sharp increase in infl ation, and mark the beginning of a large rise in the unemployment rate. In each case, both infl ation and unemployment reached a peak a few quarters after the peak in oil prices (up to a level of 11.3 percent and 13.4 percent, respectively, in the case of infl ation; 8.8 percent and 10.6 percent for the unemployment rate). The pattern of both variables during the more recent oil shock episodes is very di fferent. First, while CPI infl ation shows a slight upward trend during both O3 and O4, the magnitude of the changes involved is much smaller than that observed for O1 and O2, with the associ- ated rises in infl ation hardly standing out relative to the moderate size of fl uctuations shown by that variable since the mid- 1980s. Second, the varia- tion in the unemployment rate during and after O3 and O4 is much smaller in size than that observed in O1 and O2. The timing is also very di fferent:
Table 7.1 Postwar oil shock episodes
Run- up period 50% rise date
Max log change ($) (%)
Max log change (real)
(%)
O1 1973:3–1974:1 1974:1 104 96
O2 1979:1–1980:2 1979:3 98 85
O3 1999:1–2000:4 1999:3 91 87
O4 2002:1–2007:3 2003:1 125 110
3. While our sample ends in 2007:3, it is clear that episode (04) has not ended yet. The price of oil has continued to increase, in both 2007:4 and 2008:1.
The Macroeconomic Effects of Oil Price Shocks 381
while O1 and O2 lead to a sharp rise in unemployment, the latter variable keeps declining during the length of the O3 episode, with its rebound preced- ing O4. Furthermore, after a persistent (though relatively small) increase, unemployment starts declining in the midst of O4; that is, while the price of oil is still on the rise.
Tables 7.2 and 7.3 provide related evidence for each of the G7 coun- tries as well as for three aggregates (the G7, the euro- 12, and the OECD
Fig. 7.4 Oil shocks and CPI infl ationFig. 7.5 Oil shocks and unemployment
382 Olivier J. Blanchard and Jordi Galí
countries).
4More specifi cally, table 7.2 displays, for each country and epi- sode, the average rate of infl ation over the eight quarters following each episode’s benchmark date (at which the 50 percent threshold oil price rise is reached) minus the average rate of infl ation over the eight quarters immedi- ately preceding each run- up. Note that the increase in infl ation associated with O1 is typically larger than the one for O2. The most striking evidence, however, relates to O3 and O4, which are typically associated with a change in infl ation in their aftermath of a much smaller size than that following O1 and O2.
5The last two columns, which average the infl ation change for O1– O2 and O3– O4, makes the same point in a more dramatic way.
The evidence on output across episodes is shown in table 7.3, which reports for each country and episode (or averages of two episodes in the
Table 7.2 Oil shock episodes: Change in infl ation
O1 O2 O3 O4 AVG (1,2) AVG (3,4)
Canada 4.7 1.8 2.2 0.5 3.3 1.4
Germany 0.1 2.6 1.1 –0.2 1.4 0.4
France 5.4 3.1 1.3 0.5 4.2 0.9
U.K. 10.2 4.3 0.0 0.5 7.3 0.3
Italy 7.7 5.6 1.0 –0.1 6.6 0.4
Japan 7.9 1.0 –1.7 0.9 4.4 –0.4
U.S. 4.9 4.0 1.7 –0.2 4.5 0.7
G7 4.8 1.9 0.3 0.0 3.3 0.2
Euro12 4.3 2.7 1.3 –0.5 3.5 0.4
OECD 4.9 1.8 0.1 –0.5 3.4 –0.2
Table 7.3 Oil shock episodes: Cumulative GDP change
O1 O2 O3 O4 AVG (1,2) AVG (3,4)
Canada –8.3 –1.0 –1.5 3.2 –4.6 0.8
Germany –9.6 –3.5 1.3 –2.5 –6.6 –0.6
France –7.6 –4.4 0.6 1.2 –6.0 0.9
U.K. –16.4 –9.2 0.4 2.5 –12.8 1.4
Italy –8.6 0.4 3.0 –2.0 –4.1 0.5
Japan –16.1 –4.4 7.6 3.3 –10.3 5.4
U.S. –13.3 –11.8 –3.7 7.1 –12.5 1.7
G7 –12.6 –7.7 –0.2 3.9 –10.2 1.8
Euro12 –9.1 –2.9 1.0 –0.4 –6.0 0.3
OECD –11.2 –6.5 0.1 4.1 –8.9 2.1
4. We use quarterly data from OECD’s Economic Outlook Database. For the purpose of this exercise, infl ation is the annualized quarter- to- quarter rate of change in the CPI. These two tables have not been updated, and use data up to the end of 2005 only.
5. Even for Canada and Germany, the largest change in infl ation occurs in either O1 or O2.
The Macroeconomic Effects of Oil Price Shocks 383
case of the last two columns) the cumulative GDP gain or loss over the eight quarters following each episode’s benchmark date, relative to a trend given by the cumulative GDP growth rate over the eight quarters preceding each episode. The pattern closely resembles that shown for infl ation: O1 and O2 are generally associated with GDP losses that are much larger than those corresponding to O3 and O4 (with the latter involving some small GDP gains in some cases). When averages are taken over pairs of episodes the pattern becomes uniform, pointing once again to much larger output losses during and after the oil shocks of the 1970s.
The evidence previously presented is consistent with the hypothesis that the macroeconomic e ffects of oil price shocks have become smaller over time, being currently almost negligible (at least in comparison with their e ffects in the 1970s). But it is also consistent with the hypothesis that other (non- oil) shocks have coincided in time with the major oil shocks, either reinforcing the adverse e ffects of the latter in the 1970s, or dampening them during the more recent episodes. In order to sort out those possibilities we turn next to a more structured analysis of the comovements between oil prices and other variables.
7.3 Estimating the E ffects of Oil Price Shocks Using Structural VARs In this section we provide more structural evidence on the macroeconomic e ffects of oil price shocks, and changes over time in the nature and size of those e ffects. We provide evidence for the United States, France, Germany, the United Kingdom, Italy, and Japan, using a six- variable VAR. In the next section we turn to a more detailed analysis of the U.S. evidence, using a battery of rolling bivariate VARs.
Our baseline VAR makes use of data on the nominal price of oil (in dollars), three infl ation measures (CPI, GDP defl ator, and wages) and two quantities (GDP and employment). By using a multivariate specifi cation, we allow for a variety of shocks in addition to the oil shock that is our focus of interest. We identify oil shocks by assuming that unexpected variations in the nominal price of oil are exogenous relative to the contemporaneous values of the remaining macroeconomic variables included in the VAR. In other words, we take the oil shock to correspond to the reduced form innovation to the (log) nominal oil price, measured in U.S. dollars.
This identifi cation assumption will clearly be incorrect if economic devel-
opments in the country under consideration a ffect the world price of oil
contemporaneously. This may be either because the economy under con-
sideration is large, or because developments in the country are correlated
with world developments. For example, Rotemberg and Woodford (1996),
who rely on the same identifi cation assumption as we do when studying
the e ffects of oil shocks on the U.S. economy, restrict their sample period
to end in 1980 on the grounds that variations in the price of oil may have
384 Olivier J. Blanchard and Jordi Galí
a signifi cant endogenous component after that date. We have therefore explored an alternative assumption; namely, letting the price of oil react contemporaneously to current developments in the two quantity variables (output and employment), while assuming that quantity variables do not react contemporaneously to the price of oil. Because the contemporaneous correlations between quarterly quantity and oil price innovations are small, the results are nearly identical, and we do not report them in the text.
Another approach would be to use, either in addition or in substitution to the oil price, a more exogenous variable to proxy for oil shocks. This is the approach followed by Kilian (2008), who constructs and uses a proxy for unexpected movements in global oil production. What matters, however, to any given country is not the level of global oil production, but the price at which fi rms and households can purchase oil, which in turn depends also on world demand for oil. Thus, if the price of oil rises as a result of, say, higher Chinese demand, this is just like an exogenous oil supply shock for the remaining countries. This is indeed why we are fairly confi dent in our identifi cation approach: the large residuals in our oil price series are clearly associated either with identifi able episodes of large supply disruptions or, in the more recent past, with increases in emerging countries’ demand.
These observations largely drive our estimates and our impulse response functions.
For each of the six countries, we estimate a VAR containing six variables:
the dollar price of oil (expressed in log di fferences), CPI infl ation, GDP defl ator infl ation, wage infl ation, and the log changes in GDP and employ- ment.
6We use the dollar price of oil rather than the real price of oil to avoid dividing by an endogenous variable, the GDP defl ator. For the same reason, we do not convert the price of oil into domestic currency for non- U.S. countries. For the United States, the data are taken from the USECON database, and cover the sample period 1960:1 to 2007:3. For the remaining countries, the data are drawn from OECD’s Economic Outlook database, with the sample period being 1970:1 to 2007:3. Our three infl ation measures are quarter- to- quarter, expressed in annualized terms. Each equation in our VAR includes four lags of the six aforementioned variables, a constant term, and a quadratic trend fi tted measure of productivity growth.
Some of the oil price changes, and by implication, some of the residuals in the price of oil equation, are extremely large. The change in the price of oil for 1974:1, for example, is equal to eight times its standard deviation over the sample. Such large changes are likely to lead to small sample bias when estimating the oil price equation: the best ordinary least squares (OLS) fi t is achieved by reducing the size of these particular residuals; thus, by spuri-
6. For the United States we use nonfarm business hours instead of employment, and the wage refers to nonfarm business compensation per hour. For simplicity we use the term employ- ment to refer to both hours (in the case of the United States) and employment proper (for the remaining countries).
The Macroeconomic Effects of Oil Price Shocks 385
ously linking these very large realizations to movements in current or past values of the other variables in the regression. This in turn overstates the endogenous component of the price of oil, and understates the size of the true residuals. We deal with this issue by estimating the oil price equation using a sample that excludes all oil price changes larger than three standard deviations. (These large changes in oil prices are clearly essential in giving us precise estimates of the e ffects of oil prices on other variables. Thus, we use the complete sample when estimating the other equations.)
7.3.1 Impulse Responses
Figure 7.6, panels A through F, display the estimated impulse response functions (IRFs) for the di fferent variables of interest to an oil price shock where, as discussed previously, the latter is identifi ed as the innovation in the oil price equation. Estimates are reported for two di fferent sample peri- ods: 1970:1 to 1983:4 (1960:1 to 1983:4 for the United States) and 1984:1 to 2007:3 (1984:1 to 2005:4 for Germany and Italy). The break date chosen corresponds roughly to the beginning of the Great Moderation in the United States, as identifi ed by several authors (e.g., McConnell and Pérez- Quirós (2000). Note that each subperiod contains two of the four large oil shock episodes identifi ed in the previous section.
One standard deviation confi dence intervals, obtained using a Monte Carlo procedure, are shown on both sides of the point estimates. The esti- mated responses of GDP and employment are accumulated and shown in levels. The size of the shock is normalized so that it raises the price of oil by 10 percent on impact. This roughly corresponds to the estimated standard deviations of oil price innovations for the two subsamples, which are very similar.
7In all cases, the real price of oil shows a near- random walk response (not shown here); that is, it jumps on impact, and then stays around a new plateau.
The estimates for the United States, shown in panel A of fi gure 7.6, fi t pretty well the conventional wisdom about the e ffects of a rise in oil prices.
(fi gure 7.1, presented in the introduction, corresponds to panel A, with the results for the CPI shown in levels rather than rates of change.) For the pre- 1984 period, CPI infl ation shifts up immediately, and remains positive for a protracted period. The response of GDP infl ation and wage infl ation is similar, though more gradual. Output and employment decline persistently, albeit with a lag. Most relevant for our purposes, the responses of the same variables in the post- 1984 period are considerably more muted, thus suggest- ing a weaker impact of oil price shocks on the economy. The only exception to this pattern is given by CPI infl ation, whose response on impact is very similar across periods (though its persistence is smaller in the second period).
7. The estimated standard deviation of oil price innovations is 9.4 percent in the pre- 1984 period, 12.4 percent in the post- 1984 period.
A
B
Fig. 7.6 Impulse response to an oil price shock A, United States; B, France; C, United Kingdom; D, Germany; E, Italy; F, Japan
C
D
Fig. 7.6 (cont.)
E
Fig. 7.6 (cont.) Impulse response to an oil price shock A, United States; B, France;
C, United Kingdom; D, Germany; E, Italy; F, Japan F
The Macroeconomic Effects of Oil Price Shocks 389
This may not be surprising since part of the increase in oil prices is refl ected mechanically in the oil component of the CPI.
The estimates for France and the United Kingdom show a pattern very similar to that of the United States. In the case of France, the contrast between the early and the late periods is particularly strong, both in terms of the size and the persistence of the e ffects, and for both prices and quanti- ties. In the case of the United Kingdom, the response of infl ation variables is almost nonexistent in the latter period, though in contrast with France, there is some evidence of a decline in output and employment (albeit smaller than in the fi rst sample period).
Some of the estimated responses for Germany and Italy fi t conventional wisdom less well. The infl ation measures in Germany hardly change in response to the rise in oil prices in either period, though the impact on output and employment is more adverse in the pre- 1984 period. This is con- sistent with a stronger anti- infl ationary stance of the Bundesbank, relative to other central banks. The slight increase in employment and output in the post- 1984 period goes against conventional wisdom. In the case of Italy, there is barely any employment response in the pre- 1984 period. Still, for both countries the sign of most of the responses accord with conventional wisdom, and the responses are smaller in the post- 1984 period.
The story is di fferent for Japan. The sign of many of the responses to the rise in oil prices is often at odds with standard priors. Also, the uncertainty of the estimates is much larger, as refl ected in the wider bands. The e ffect on infl ation is weak and does not have a clear sign in either period. There is a (slight) rise in output in both periods, and of employment in the post- 1984 period.
In short, except for Japan (and to some extent, for Germany), most of the responses fi t conventional wisdom rather well: an increase in the price of oil leads to more wage and price infl ation, and to a decrease in employment and output for some time. In all cases, however, the e ffects on both infl a- tion and activity are considerably weaker in the second subsample than in the fi rst.
7.3.2 Variance and Historical Decompositions
How important are oil shocks in accounting for the observed fl uctua-
tions in infl ation, output, and employment in the U.S. economy? Table 7.4
and fi gure 7.7 answer this question by using the decomposition associated
with the estimated six- variable VAR, with data starting in 1960. For each
variable and sample period, they compare the actual time series with the
component of the series that results from putting all shocks, except the
identifi ed oil price shocks, equal to zero. Series for GDP and employment
are accumulated, so the resulting series are in log- levels. All series are then
Hodrick- Prescott (HP)- fi ltered so that the series can be interpreted as devia-
tions from a slowly moving trend. Table 7.4 provides statistics for the role
390 Olivier J. Blanchard and Jordi Galí
of oil shocks as a source of fl uctuations, including its percent contribution to the volatility of each variable (including the real price of oil, measured relative to the GDP defl ator), both in absolute and relative terms. Figure 7.7 plots the series over time.
The estimated standard deviations of the oil- driven component of the di fferent variables (“conditional standard deviations”), given in the fi rst three columns of table 7.4, show that the volatility of fl uctuations caused by oil shocks has diminished considerably for all variables, except for the real price of oil itself. In fact, the standard deviation of the exogenous com-
Table 7.4 The contribution of oil shocks to U.S. economic fl uctuations, 1960:1–2007:3
Conditional standard deviation
Conditional SD Unconditional SD
60:1–83:4 84:1–07:3 Ratio 60:1–83:4 84:1–07:3
Oil price (real) 12.9 15.4 1.19 0.82 0.88
CPI infl ation 0.89 0.74 0.83 0.43 0.55
GDP infl ation 0.71 0.15 0.24 0.50 0.25
Wage infl ation 0.69 0.56 0.81 0.41 0.23
GDP 0.59 0.28 0.48 0.34 0.31
Hours 0.76 0.43 0.57 0.42 0.30
Fig. 7.7 The role of oil price shocks
The Macroeconomic Effects of Oil Price Shocks 391
ponent of the latter variable is about 20 percent larger in the second sample period. This can be explained to a large extent by the limited variation in the real price of oil before the 1973 crisis, and despite the two large spikes in that year and during 1979 and 1980.
This evidence reinforces our earlier Impulse response function (IRFs)- based fi ndings of a more muted response of all variables to an oil shock of a given size. Thus, the change in the way the economy has responded to oil shocks has contributed to the dampening of economic fl uctuations since the mid- 1980s, the phenomenon known as the Great Moderation. Interestingly, our estimates suggest that this has been possible in spite of the slightly larger volatility of oil prices themselves.
The next two columns of table 7.4 give the relative contribution of oil shocks to movements in the various variables, measured as the ratio of the conditional to the unconditional standard deviation. The estimates suggest that the relative contribution of oil shocks to fl uctuations in quantity vari- ables (GDP and employment) has remained roughly unchanged over time, at around one- third. In the case of wage infl ation and GDP defl ator infl ation, the contribution of oil shocks has declined to one- fourth in both cases, from a level close to one- half. In contrast, the contribution of oil shocks to CPI infl ation has increased in the recent period. Note that this is consistent with a relatively stable core CPI, with oil price changes being passed through to the energy component of the CPI, and accounting for, according to our esti- mates, as much as 60 percent of the fl uctuations in overall CPI infl ation.
Figure 7.7 allows us to focus on the contribution of oil prices to the 1973 to 1974 and 1979 to 1981 episodes. It shows the substantial but nonexclusive role of exogenous oil shocks during each of the two episodes. In particu- lar, while for our three infl ation variables the oil price shocks seem to have accounted for the bulk of the increases in 1973 to 1974 and 1979 to 1981, no more than a half of the observed decline in employment and output during those episodes can be attributed to the oil shocks themselves. Thus, our fi ndings suggest that other shocks played an important role in triggering those episodes.
Within our six- variable VAR, our partial identifi cation approach does not allow us to determine what those additional underlying shocks may have been. Yet when we replace the price of oil by the broader producer price index (PPI) for crude materials in our six- variable VAR, the estimates of GDP and employment driven by exogenous shocks to that broader price index track more closely the movements of the actual time series themselves in the pre- 1984 period, including the two large oil shock episodes contained in that period, as shown in fi gure 7.8. In particular, those shocks account for more than half of the fl uctuations in all variables over the pre- 1984 period.
On the other hand, such broader supply shocks play a very limited role in
accounting for the fl uctuations in output and employment in the post- 1984
period (though they play a more important one in accounting for variations
in CPI infl ation, in a way consistent with earlier evidence).
392 Olivier J. Blanchard and Jordi Galí
7.4 U.S. Evidence Based on Rolling Bivariate Regressions
So far, we have analyzed the macroeconomic e ffects of oil price shocks and their change over time under the maintained assumption of a discrete break sometime around the mid- 1980s. While the fi ndings reported previ- ously are largely robust to changes in the specifi c date of the break, some of the potential explanations (discussed following) for the change in the e ffects of oil price shocks are more likely to have been associated with a more gradual variation over time. This leads us to adopt a more fl exible approach, and estimate rolling IRFs to oil price shocks, based on a simple dynamic equation linking a variable of interest to its own lags and the current and lagged values of the change in the (log) oil price. We do this using a moving window of 40 quarters, with the fi rst moving window centered in 1970.
More specifi cally, letting y
tand p
todenote the variable of interest and the price of oil, respectively, we use OLS to estimate the regression:
y
t∑j14
j
y
tj∑j04
j
p
otju
tand use the resulting estimates to obtain the implied dynamic response of y
t(or a transformation thereof) to a permanent 10 percent (log) change in the price of oil, thus implicitly assuming in the simulation that p
tois an i.i.d.
process (which is roughly consistent with the random walk- like response of the price of oil obtained using our multivariate model).
Fig. 7.8 The role of shocks to crude materials prices
The Macroeconomic Effects of Oil Price Shocks 393
Relative to the multivariate model analyzed in the previous section, cor- rect identifi cation of oil price shocks is obviously more doubtful in the present bivariate model, given the lower dimension specifi cation of the economy’s dynamics. This shortcoming must be traded- o ff with the pos- sibility of estimating the VAR with much shorter samples and, hence, being able to obtain our rolling IRFs. In order to check the consistency with our earlier results, we fi rst computed the average IRFs across moving windows within each of the subperiods considered earlier (pre- 1984 and post- 1984), and found the estimated IRFs (not shown) to be very similar to the ones obtained earlier. In particular, both the infl ation variables, as well as output and employment, show a more muted response in the more recent period.
Figure 7.9, panels A through E, display the rolling IRFs for our three infl ation measures, output, and employment. Several features stand out in the fi gure.
Consumer price index infl ation appears quite sensitive to the oil shock over the entire sample period, but particularly in the late 1970s, when infl a- tion is estimated to rise more than 1 percentage point two/ three quarters after a 10 percent rise in the oil price. The response becomes steadily more muted over time and, perhaps as important, less persistent, especially in the more recent period (in a way consistent with our earlier evidence based on the six- variable VAR). The evolution over time in the response of GDP defl ator infl ation to an oil price shock is similar to that of CPI infl ation, but shows a more dramatic contrast, with the response at the end of our sample being almost negligible. The response of wage infl ation is rather muted all along, except for its large persistent increases in the late 1970s and early 1980s, and a similar spike in the 1990s.
The most dramatic changes are in the responses of output and employ- ment (see fi gure 7.9, panels D and E). In the early part of the sample, output is estimated to decline as much as 1 percent two years after the 10 per- cent change in the price of oil. The estimated response, however, becomes weaker over time, with the point estimates of that response becoming slightly positive for the most recent period. A similar pattern can be observed for employment.
The previous evidence thus reinforces the picture that emerged from the earlier evidence, one which strongly suggests a vanishing e ffect of oil shocks on macroeconomic variables, both real and nominal. In the next section we try to uncover some of the reasons why.
7.5 Modeling the Macroeconomic E ffects of Oil Price Shocks: A Simple Framework
We now develop a simple model of the macroeconomic e ffects of oil price
shocks. Our focus is on explaining the di fferent response of the economy to
oil price shocks in the 1970s and the 2000s. With this in mind, we focus on
three potential changes in the economy.
A
B
Fig. 7.9 Response of infl ation, GDP, and employment
C
D
Fig. 7.9 (cont.)
396 Olivier J. Blanchard and Jordi Galí
First, the behavior of wages. To us, this looks a priori like the most plausible candidate. The 1970s were times of strong unions and high wage indexation. In the 2000s, unions are much weaker, and wage indexation has practically disappeared.
The second potential change is the role of monetary policy. Faced with a new type of shock, the central banks of the 1970s did not know at fi rst how to react, policy mistakes were made, and central bank credibility was low. In the 2000s, supply shocks are no longer new, monetary policy is clearly set, and credibility is much higher.
Third, and trivially, is the quantitative importance of oil in the economy.
Increases in the price of oil have led to substitution away from oil, and a decrease in the relevant shares of oil in consumption and in production. The question is whether this decrease can account for much of the di fference in the e ffects of oil prices in the 1970s and the 2000s.
8We start from the standard new- Keynesian model and introduce two modifi cations. First, we introduce oil both as an input in consumption and as an input in production. We assume the country is an oil importer, and that
E
Fig. 7.9 (cont.)
8. Some observers have suggested another factor—an increase in hedging against oil price shocks by oil users. What is known about hedging by airlines suggests, however, that while hedging is more prevalent than in the 1970s, its extent remains limited, with few hedges going beyond a year. See, for example, Carter, Rogers, and Simkins (2006a, 2006b).
The Macroeconomic Effects of Oil Price Shocks 397
the real price of oil (in terms of domestic goods) follows an exogenous pro- cess. Second, we allow for real wage rigidities, along the lines of our earlier work (Blanchard and Galí 2007). We present only log- linearized relations in the text, leaving the full derivation to appendix A. Lower case letters denote logarithms of the original variables, and for notational simplicity, we ignore all constants.
7.5.1 The Role of Oil
Oil is used both by fi rms in production and by consumers in consumption.
Production is given by
q
ta
tn
n
tm
m
t,
where q
tis (gross) domestic output; a
tis an exogenous technology parameter;
n
tis labor; m
tis the quantity of imported oil used in production; and
nm
1.
9Consumption is given by
c
t⬅ (1 ) c
q,tc
m,t,
where c
tis consumption; c
q,tis the consumption of domestically produced goods (gross output); and c
m,tis the consumption of imported oil.
In this environment, it is important to distinguish between two prices, the price of domestic output p
q,t, and the price of consumption p
c,t. Let p
m,tbe the price of oil, and s
t⬅ p
m,t– p
q,tbe the real price of oil. From the defi nition of consumption, the relation between the consumption price and the domestic output price is given by
(1) p
c,tp
q,ts
t.
Increases in the real price of oil lead to an increase in the consumption price relative to the domestic output price.
7.5.2 Households
The behavior of households is characterized by two equations. The fi rst is an intertemporal condition for consumption:
(2) c
tE
t{c
t1} (i
tE
t{
c,t1}),
where i
tis the nominal interest rate, and
c,t⬅ p
c,t– p
c,t– 1is CPI infl ation.
The second condition characterizes labor supply. If the labor market was perfectly competitive, labor supply would be implicitly given by
w
tp
c,tc
tn
t,
9. We use a Cobb- Douglas specifi cation for convenience. It has the counterfactual implica- tion that the share of oil in output remains constant. So, in our framework, when looking at changes in the share over time, we must attribute it to a change in the parameter m. For our purposes, this appears innocuous.
398 Olivier J. Blanchard and Jordi Galí
where w
tis the nominal wage, and n
tis employment. This is the condition that the consumption wage must equal the marginal rate of substitution between consumption and leisure; is the inverse of the Frisch elasticity of labor supply.
We formalize real wage rigidities by modifying the previous equation to read
(3) w
tp
c,t(1 ) (c
tn
t),
where we interpret the parameter ∈ [0, 1] as an index of the degree of real wage rigidities. While clearly ad- hoc, equation (3) is meant to cap- ture in a parsimonious way the notion that real wages may not respond to labor market conditions as much as implied by the model with perfectly competitive markets. We have explored the implications of a dynamic ver- sion of equation (3), in which the wage adjusts over time to the marginal rate of substitution. This alternative is more attractive conceptually, and gives richer dynamics. However, it is also analytically more complex, and we have decided to present results using the simpler version presented earlier.
7.5.3 Firms
Given the production function, cost minimization implies that the fi rms’
demand for oil is given by m
t–
tp– s
tq
t, where
tpis the price markup.
Using this expression to eliminate m
tin the production function gives a reduced- form production function
(4) q
t1
1
m(a
tn
n
tm
s
tm
tp
).
Output is a decreasing function of the real price of oil, given employment and technology.
Combining the cost minimization conditions for oil and for labor with the aggregate production function yields the following factor price frontier:
(5) (1
m) (w
tp
c,t) (
m(1
m) ) s
t(1
nm
) n
ta
ttp
0.
Given productivity, an increase in the real price of oil must lead to one
or more of the following adjustments: (a) a lower consumption wage, (b)
lower employment, and (c) a lower markup. Under our assumed functional
forms, it can be shown that with fl exible prices and wages, the entire burden
of the adjustment in response to an increase in s
tfalls on the consumption
wage, with employment and the markup remaining unchanged. But, as we
discuss next, things are di fferent when we allow the markup to vary (as a
The Macroeconomic Effects of Oil Price Shocks 399
result of sticky prices), and wages to respond less than their competitive labor markets counterpart.
Firms are assumed to set prices à la Calvo (1983), an assumption that yields the following log- linearized equation for domestic output price infl a- tion (domestic infl ation for short)
(6)
q,tE
t{
q,t1}
ptp
,
where
p⬅ [(1 – )(1 – )/ ][(
mn
)/ (1 (1 –
mn
)( ε – 1))], where denotes the fraction of fi rms that leave prices unchanged, is the discount factor of households, and ε is the elasticity of substitution between domestic goods in consumption.
Note that this specifi cation assumes a constant desired markup of fi rms. By doing so, we rule out a mechanism examined by Rotemberg and Woodford (1996) who argue that, to explain the size of the decline in out- put observed in response to oil shocks, one must assume countercyclical markups. We do so not because we believe the mechanism is irrelevant, but because we do not think that variations in the degree of countercyclicality of markups are likely to be one of the main factors behind the di fferences between the 1970s and the 2000s.
7.5.4 Equilibrium
The real wage consistent with household choices (cum real wage rigidi- ties) is given by equation (3), and depends on consumption and employ- ment.
The real wage consistent with the fi rms’ factor price frontier is given by equation (5) and depends on the real price of oil, the markup, and employ- ment.
Together, these two relations imply that the markup is a function of con- sumption, employment, and the real price of oil. Solving for consumption by using the condition that trade be balanced gives:
(7) c
tq
ts
ttp
,
where ⬅
m/ ( M
p–
m), with M
pdenoting the steady- state gross markup (now in levels). Combining this equation with the reduced- form production function gives consumption as a function of employment, productivity, the real price of oil, and the markup
c
t1
1
ma
tn
1
mn
t冢 1
m m冣 st 冢 1
m m冣
tp.
.
If the steady- state markup is not too large, the last term is small and can
safely be ignored. Replacing the expression for consumption in equation (3)
for the consumption wage and then replacing the consumption wage in the
factor price frontier gives an expression for the markup
400 Olivier J. Blanchard and Jordi Galí
(8)
tpn
n
ts
s
ta
a
t, where
n⬅ (1
nm
) (1
m)(1 )(1 )
1 (1 )(
m(1
m) ) 0
a⬅
1 (1 )(
m(1
m) ) 0
s⬅ (
m(1
m) )
1 (1 )(
m(1
m) ) 0.
Using this expression for the markup in equations (6) and (2) gives the following characterization of domestic infl ation
(9)
q,tE
t{
q,t1}
pnn
tps
s
tpa
a
t.
Under our assumptions, the fi rst best level of employment can be shown to be invariant to the real price of oil—substitution and income e ffects cancel.
10If 0; that is, if there are no real wage rigidities, then
aand
sare both equal to zero, and domestic infl ation only depends on employment. Together, these two propositions imply that stabilizing domestic infl ation is equivalent to stabilizing the distance of employ- ment from fi rst best—a result we have called elsewhere the “divine co- incidence.”
Positive values of lead instead to positive values of
aand
s. The higher , or the higher (
m(1 –
m) )—an expression that depends on the shares of oil in production and in consumption—the worse the trade- o ff between stabilization of employment and stabilization of domestic infl ation in response to oil price shocks.
7.5.5 Implications for GDP and the GDP Defl ator
Note that the characterization of the equilibrium did not require intro- ducing either value added or the value- added defl ator. But these are needed to compare the implications of the model to the data.
The value- added defl ator, p
y,t, is implicitly defi ned by p
q,t(1 –
m) p
y,tm
p
m,t. Rearranging terms gives
(10) p
y,tp
q,tm
1
ms
t,
thus implying a negative e ffect of the real price of oil on the value- added defl ator, given domestic output prices.
10. To see this, we can just determine equilibrium employment under perfect competition in both goods and labor markets, corresponding to the assumptions t 0 for all t and 0, respectively.
The Macroeconomic Effects of Oil Price Shocks 401
The defi nition of value added, combined with the demand for oil, yields the following relation between value added and output:
(11) y
tq
tm
1
ms
ttp
.
This in turn implies the following relation between value added and con- sumption:
(12) y
tc
t冢 1
m m冣 s
t.
An increase in the price of oil decreases consumption given value added both because (imported) oil is used as an input in production, and used as an input in consumption.
Under the same approximation as before; that is, ( –
m/ (1 –
m))
tp⯝ 0, equations (4) and (11) imply the following relation between value added and employment:
(13) y
t1
1
m(a
tn
n
t).
Note that, under this approximation, the relation between value added and employment does not depend on the real price of oil.
7.5.6 Quantifying the E ffects of Oil Price Shocks
Equations (1), (2), (9), (12), and (13) describe the equilibrium dynamics of prices and quantities, given exogenous processes for technology and the real price of oil, and a description of how the interest rate is determined (i.e., an interest rate rule). We now use these conditions to characterize the economy’s response to an oil price shock.
Assume that a
t0 for all t (i.e., abstract from technology shocks). It fol- lows from (13) and the previous discussion that the e fficient level of value added is constant (and normalized to zero) in this case. Assume further that the real price of oil follows an AR(1) process
(14) s
ts
s
t1ε
t.
We can then summarize the equilibrium dynamics of value added and domestic infl ation through the system:
(15)
q,tE
t{
q,t1} κ y
tps
s
t(16) y
tE
t{y
t1} (i
tE
t{
q,t1})
m(1
s)
1
ms
t,
where κ ⬅
pn(1 –
m)/
n.
402 Olivier J. Blanchard and Jordi Galí